aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp
diff options
context:
space:
mode:
authorGeorgios Pinitas <georgios.pinitas@arm.com>2020-11-17 18:46:40 +0000
committerMichele Di Giorgio <michele.digiorgio@arm.com>2020-11-23 13:50:40 +0000
commit573c7bbb15c8f205e93397105e66fbc5102882e4 (patch)
tree00f34dfe387f3a264e6eb74d0665b2190ab4d3bb /src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp
parent34c0c8a7d841111a0147ef84e29627f3e610d273 (diff)
downloadComputeLibrary-573c7bbb15c8f205e93397105e66fbc5102882e4.tar.gz
Update tuning numbers for A55 for both fp16 and fp32
Resolves: COMPMID-3974 Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com> Change-Id: I6d5189e44ebeda1575a80dd14ec3a09c75f68e03 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4521 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> (cherry picked from commit 40943df83026b66356f24e30f31f78c8b9e59c92) Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4169 Tested-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp12
1 files changed, 12 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp
index 73fb5b7122..24e258e4b8 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hgemm_8x24.hpp
@@ -25,6 +25,7 @@
#if defined(__aarch64__) && (defined(FP16_KERNELS) || defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC))
+#include "../performance_parameters.hpp"
#include "../std_transforms_fixed.hpp"
namespace arm_gemm {
@@ -61,6 +62,17 @@ public:
// Use the standard fixed size transforms.
StdTransformsFixed<operand_type, result_type, 8, 24> transforms = {};
+ static PerformanceParameters get_performance_parameters(const CPUInfo *ci)
+ {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A55r1:
+ return { 7.16, 1.14, 0.67 };
+
+ default:
+ return { 12.67, 3.98, 1.16 };
+ }
+ }
+
// Default to the generic kernel
kern_type kernel = a64_hgemm_asimd_8x24;